Data-driven Causal Modeling And Analysis Of Traffic Incident

The rapid development of big data technology has brought opportunities and challenges to the research in the field of intelligent transportation,and also provides strong support for the causality research on the impact of freeway traffic incidents.Based on the incident feature,the propagation characteristics of incidents,time variability of traffic status and causal relevance of road sections,a prediction model for the propagation of the traffic incident on freeways is established in this paper.Through the combination of big data and traffic knowledge,a method is proposed to quantify the traffic incident impact and build a causal mechanism of the traffic flow changes on the freeway network after the occurrence of incident,considering the causality of incident.The main contents and innovations of this paper are as follows:(1)Aiming at the causality research of traffic incidents,the incident impact on the traffic flow of the freeway network is analyzed,and the propagation law of the incident impact is explored.Firstly,comparing the changes and sensitivity of traffic flow parameters under the influence of traffic incident,it is clear that speed is an important parameter for detecting the impact.Then,the speed change ratio is selected as the index to characterize and identify the effect of incidents.Finally,this paper proposes a novel method to model the incident impact as defining the propagation characteristic curve to determine a time-varying spatial area by the spatiotemporal diagram of the speed change ratio.(2)For the mining of factors affecting the spread of traffic incidents,Boruta algorithm are used to obtain the set of relevant features that are maximally correlated with traffic incident impact.The research results show that there is a certain correlation between feature sets of traffic incidents,including spatiotemporal characteristics,environmental characteristics,and supplementary characteristics.The main factors have been screened out,which affect the evolution of the road network traffic state under the influence of traffic incidents.(3)Based on Rete algorithm,a design method of traffic incident rule engine is proposed to predict the propagation of traffic incidents on the freeway network.On the basis of traffic incident feature subsets,traffic incidents are classified to establish incident impact rules and obtain the influence laws of various traffic incidents in the traffic flow of the network.For reusing and sharing domain knowledge of traffic incidents,the freeway traffic incident knowledge graph is constructed based on the ontology model.The research results show that a prediction model for the propagation of the traffic incident by combining rule-based reasoning with knowledge graphs can better solve the problem of quantifying the impact.At the same time,the process with the propagation of incident impact also can be comprehensively explained.(4)Based on the Bayesian framework,a method for the propagation of traffic incident impact is proposed.The study on the causality of traffic incidents is understood as exploring the propagation process of deceleration with causal effects under the influence of traffic incidents.First,a data-driven approach is proposed to partition historical traffic flow data sets with traffic incident features and establish the pattern of incidents.Based on the Bayesian additive regression tree,a prediction model is established to infer the incident impact area,and the average causal effect is construct to quantify the impact of deceleration.This paper further demonstrates the proposed method in a case study by taking advantage of an incident record and related real freeway traffic data.